AM2iCo: Evaluating Word Meaning in Context across Low-Resource Languages with Adversarial Examples